A non-centred reparameterization may help more when there’s no much data, but sometimes it may be required (anyway) to achieve convergence if the geometry of the posterior is too nasty, the stan user manual has a good entry explaining the details: Stan User’s Guide
Regarding your model, I see a big issue. You are adding an extremely large number of variables to the expectation, some forming interactions as well. I didn’t check your model in detail, but unless you have very good clarity on how these variable relate (causally) to each other, you may induce unwanted spurious effects (e.g. as induced by a collider). You may want to check the McElreath’s Rethinking book (chapter 6 in the second edition, if I’m correct, or lecture 6 here). This paper is also a great source for getting a notion on this topic: https://ftp.cs.ucla.edu/pub/stat_ser/r493.pdf
Personally, I’d try to build up the model incrementally (even starting from simulated data), to understand whether it’s actually answering the question I want to ask. Maybe you can start with some prior predictive checks to better understand and tune your priors, if you don’t have good prior information that can help you to define them. See here: Prior and Posterior Predictive Checks — PyMC 4.1.4 documentation
I hope this helps a bit.